WO2020043041A1

WO2020043041A1 - Method and device for point cloud data partitioning, storage medium, and electronic device

Info

Publication number: WO2020043041A1
Application number: PCT/CN2019/102486
Authority: WO
Inventors: 曾超
Original assignee: 腾讯科技（深圳）有限公司
Priority date: 2018-08-27
Filing date: 2019-08-26
Publication date: 2020-03-05
Also published as: CN110148144A; CN110148144B; US11282210B2; US20200410690A1

Abstract

Disclosed in the present application are a method and device for point cloud data partitioning, a storage medium, and an electronic device. The method comprises: acquiring target point cloud data, wherein the target point cloud data is data obtained by using a laser beam to scan target objects surrounding a vehicle; clustering the target point cloud data to obtain a plurality of first data sets, wherein feature points represented by the point cloud data comprised in each first data set is fit on the same partitioning segment, and the feature points are points on the target objects; merging the plurality of first data sets according to the distance between the plurality of partitioning segments to obtain second data sets, wherein the second data sets comprise at least one first data set. The present application solves the technical problem of inefficient point cloud partitioning in the related art.

Description

Method and device for segmenting point cloud data, storage medium, and electronic device

This application claims the priority of a Chinese patent application filed on August 27, 2018 with the Chinese Patent Office, application number 201810982858.0, and application name "Segmentation Method and Device for Point Cloud Data, Storage Medium, Electronic Device". Citations are incorporated in this application.

Technical field

This application relates to the field of autonomous driving, and in particular, to segmentation technology of point cloud data.

Background technique

For the 3D reconstruction of large scenes, it has received great attention due to its important applications in 3D city maps, road maintenance, urban planning, and autonomous driving. Depth sensors and position and attitude sensors are used to collect three-dimensional information of the surrounding environment based on fixed stations or mobile platforms. They are widely used due to their high-efficiency, real-time, and high-precision characteristics. Because the scanned scene may contain different objects, such as the ground, buildings, trees, vehicles, etc., before performing 3D reconstruction, the point cloud data belonging to different objects need to be separated from each other by point cloud segmentation so that each object can be separated separately. Perform point cloud modeling.

Point cloud segmentation algorithms in related technologies need to scan the point cloud data multiple times, which is computationally expensive and inefficient, and does not meet real-time processing requirements.

In view of the above problems, no effective solution has been proposed.

Summary of the Invention

The embodiments of the present application provide a method and device for segmenting point cloud data, a storage medium, and an electronic device, so as to at least solve the technical problem of low efficiency of point cloud segmentation in related technologies.

According to an aspect of the embodiment of the present application, a method for segmenting point cloud data is provided, including: obtaining target point cloud data, where the target point cloud data is data obtained by scanning a target object around a vehicle through a laser beam; Clustering the target point cloud data to obtain multiple first data sets, wherein the feature points represented by the point cloud data included in each first data set are fitted to the same segmentation line segment, and the feature points are on the target object Point; combining a plurality of first data sets according to the distance between the plurality of divided line segments to obtain a second data set, wherein the second data set includes at least one first data set.

According to another aspect of the embodiments of the present application, a point cloud data segmentation device is further provided, including: an obtaining unit for obtaining target point cloud data, where the target point cloud data is a target around a vehicle through a laser wire harness Data obtained by scanning objects; a clustering unit configured to cluster target point cloud data to obtain a plurality of first data sets, wherein feature points represented by the point cloud data included in each first data set are fitted On the same segmentation line segment, feature points are points on the target object; a merging unit is configured to merge a plurality of first data sets according to the distance between the plurality of segmentation line segments to obtain a second data set, where: The second data set includes at least one first data set.

According to another aspect of the embodiments of the present application, a storage medium is also provided. The storage medium includes a stored program, and the method is executed when the program is run.

According to another aspect of the embodiments of the present application, an electronic device is further provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the foregoing method through the computer program.

According to another aspect of the embodiments of the present application, there is also provided a computer program product including instructions, which, when run on a computer, cause the computer to execute the above method.

In the embodiment of the present application, target point cloud data is obtained, and the target point cloud data is data obtained by scanning a target object around a vehicle through a laser beam; the target point cloud data is clustered to obtain a plurality of first data sets, each The feature points represented by the point cloud data included in each of the first data sets are fitted to the same segmentation line segment, and the feature points are points on the target object; multiple first data sets are compared according to the distance between the plurality of segmentation line segments. The merging is performed to obtain a second data set, and the second data set includes at least one first data set. In the segmentation process, "clustering the target point cloud data to obtain multiple first data sets" is equivalent to completing the point cloud data segmentation only after traversing all the point cloud data once, instead of using multiple Iterating over the point cloud data to complete the segmentation can solve the technical problem of low efficiency of point cloud segmentation in related technologies, and then achieve the technical effect of improving the segmentation efficiency.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings described here are used to provide a further understanding of the present application and constitute a part of the present application. The schematic embodiments of the present application and the description thereof are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

1 is a schematic diagram of a hardware environment of a method for segmenting point cloud data according to an embodiment of the present application;

2 is a flowchart of an optional method for segmenting point cloud data according to an embodiment of the present application;

3 is a schematic diagram of an optional lidar according to an embodiment of the present application;

4 is a schematic diagram of an optional point cloud data according to an embodiment of the present application;

5 is a schematic diagram of an optional lidar scene according to an embodiment of the present application;

6 is a flowchart of an optional method for segmenting point cloud data according to an embodiment of the present application;

7 is a schematic diagram of an optional adaptive distance threshold according to an embodiment of the present application;

8 is a schematic diagram of an optional segmentation of point cloud data according to an embodiment of the present application;

9 is a schematic diagram of an optional device for segmenting point cloud data according to an embodiment of the present application;

FIG. 10 is a structural block diagram of a terminal according to an embodiment of the present application.

detailed description

In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only Examples are part of this application, but not all examples. Based on the embodiments in this application, all other embodiments obtained by a person of ordinary skill in the art without creative efforts should fall within the protection scope of this application.

It should be noted that the terms “first” and “second” in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data so used may be interchanged where appropriate so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. Furthermore, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units need not be limited to those explicitly listed Those steps or units may instead include other steps or units not explicitly listed or inherent to these processes, methods, products or equipment.

First, some terms or terms appearing during the description of the embodiments of the present application are applicable to the following explanations:

Autonomous vehicles (Self-piloting vehicles), also known as self-driving cars, computer-driven cars, or wheeled mobile robots, are smart cars that realize driverlessness through computer systems.

High-definition maps are high-precision, fine-definition maps that require decimeter accuracy to be able to distinguish individual lanes. Today, with the development of positioning technology, high-precision positioning has become possible. The refined definition requires formatting and storage of various traffic elements in the traffic scene, including road map data, lane network data, lane lines, and traffic signs in traditional maps.

According to an aspect of the embodiments of the present application, a method embodiment of a method for segmenting point cloud data is provided.

Optionally, in this embodiment, the foregoing method for segmenting point cloud data may be applied to a hardware environment composed of a server 101 and / or a terminal 103 as shown in FIG. 1. As shown in FIG. 1, the server 101 is connected to the terminal 103 through the network, and can be used to provide services (such as game services, application services, map services, autonomous driving, etc.) to the terminal or the client installed on the terminal. Or, a database 105 is provided independently of the server 101 to provide data storage services for the server 101. The above networks include, but are not limited to: a wide area network, a metropolitan area network, or a local area network. The terminal 103 is a smart terminal that can be used in a vehicle, and is not limited to Car equipment, mobile phones, tablets, etc.

The method for segmenting point cloud data in the embodiment of the present application may be performed by the server 101. FIG. 2 is a flowchart of an optional method for segmenting point cloud data according to the embodiment of the present application. As shown in FIG. 2, the method It can include the following steps:

Step S202: The server obtains target point cloud data.

The target point cloud data is data obtained by scanning a target object around the vehicle with a laser beam. In an implementation manner, the above-mentioned target point cloud data may be data obtained by scanning multiple laser beams of a lidar. The lidar may measure the size and shape of an object by using a scanning technique, and the lidar may adopt a stability And a precision rotating motor, when the laser beam hits the polygonal gauge driven by the motor to form a scanning beam, because the polygonal gauge is located on the front focal plane of the scanning lens, and the laser beam is evenly rotated to the mirror The incident angle changes relatively continuously, so the reflection angle also changes continuously. Through the function of the scanning lens, a parallel and continuous top-to-bottom scanning line is formed to form the scanning line data, that is, a single-beam laser scans once Sequence of formed point clouds.

The lidar of the present application may be a low-beam lidar or a multi-beam lidar. A low-beam lidar scan can generate fewer line-beam scan lines at a time. A low-beam lidar generally includes a 4-beam and an 8-beam, mainly 2.5D lidar The vertical field of view generally does not exceed 10 °; multi-line beam lidar (or 3D lidar) can generate multiple scan lines in one scan. Multi-line beam lidar generally includes 16 line beams, 32 line beams, 64 line beams, etc. 3D lidar and 2.5 The biggest difference between D lidar lies in the range of vertical field of view of lidar. The range of vertical field of view of 3D laser mine can reach 30 ° or even more than 40 °.

In the Advanced Driver Assistance System ADAS (Advanced Driver Assistance System), it is necessary to use a variety of sensors installed on the car to collect the environmental data inside and outside the car at the first time to identify and detect static and dynamic objects. Active safety technology that enables drivers to detect possible dangers in the fastest time, so as to attract attention and improve safety. The sensors used by ADAS are mainly cameras, lidars, etc. When the vehicle detects potential danger, it will issue an alert to remind the driver to pay attention to abnormal vehicle or road conditions. In this case, the target object can be static and dynamic outside the vehicle. Objects, such as buildings, pedestrians, other vehicles, animals, traffic lights, etc., are used to determine whether they are potential dangers and assist driving logic.

Step S204: The server clusters the target point cloud data to obtain a plurality of first data sets.

The feature points represented by the point cloud data included in each first data set are fitted to the same segmentation line segment, that is, the feature points represented by the point cloud data stored in the first data set are located at the positions corresponding to the first data set. The feature points on the segment are the points on the target object. It is equivalent to a segmentation method based on each scan line. Each scan line is segmented to obtain a segment line segment on each scan line, so that all point cloud data is initially segmented, such as obtaining a dataset of the boundary of the target object, and the target object. Appearance dataset, ground dataset, etc.

In step S206, the server merges the plurality of first data sets according to the distance between the plurality of divided line segments to obtain a second data set.

Each second data set includes at least one first data set, which is equivalent to using a plane scanning method to merge the scan line segmentation line segments to obtain a candidate target clustering set. After obtaining the candidate target clustering set, feature extraction can be performed on the candidate target clustering set, noise and ground sets are eliminated, and the final segmentation result is obtained, that is, the second data set and the merged segmentation line segment.

After analyzing the related technology, the applicant realized that most of the point cloud segmentation methods in the related technology deal with unordered and discrete point cloud data. Among the point cloud segmentation methods, the clustering segmentation method has low complexity due to its method It is easy to implement and can be used for the segmentation of spatial point clouds. However, due to the existence of ground point cloud data in large outdoor scenes, it is difficult to effectively segment ground and non-ground objects using clustering methods.

In an optional embodiment, in the non-ground point cloud clustering and segmentation part, a clustering segmentation method based on a fixed threshold radius may be adopted. The selection of the threshold value has a great influence on the segmentation result. If the threshold value is too large, the separation distance Closer small objects may not be separated as one object (that is, under segmentation). If the threshold is selected too small, objects with a large separation distance (such as buildings) may be divided into different objects (that is, over segmentation occurs). ).

In the embodiment of the present application, first, based on the segmentation method of each scan line, segment line segments (may be referred to as segment segments) on each scan line are obtained, and each segment line segment corresponds to a first data set, which can implement object The initial segmentation of the contours; then the plane scanning method is used to merge the segmented segments of each scan line to obtain the candidate target clustering set (that is, the second data set), which is equivalent to merging the contour lines that belong to an object object. The target point cloud data needs to be traversed once, which can solve the problem of low efficiency and difficult to effectively segment ground and non-ground objects when using only clustering and segmentation. In addition, because the solution of this application only needs to determine the distance between segmented lines, relative to the distance between point clouds, there is no accidental factor for determining the distance between point clouds (some point clouds overlap in a certain plane dimension due to the perspective, but actually exist in It does not overlap in space), and can also avoid the problem of under-segmentation or over-segmentation that occurs when the point cloud is directly segmented using a clustering segmentation method based on a fixed threshold radius. It can be seen that, while reducing the complexity of the algorithm, the technical solution of the present application is beneficial to reduce the over-segmentation and under-segmentation in the segmentation process, and improve the perceived robustness of autonomous driving.

The foregoing embodiment takes the point cloud data segmentation method of the present application as an example for description by the server 101. The point cloud data segmentation method of the present application can also be executed by the terminal 103. The difference from the above embodiment lies in the execution subject. Converted by server to terminal. The point cloud data segmentation method of the present application may also be performed by the server 101 and the terminal 103 jointly, and one or two of the steps (such as step S202) are performed by the terminal 103, and the remaining steps (such as step S204-step S206) are performed by the server 101 ). The method for segmenting point cloud data performed by the terminal 103 in the embodiment of the present application may also be performed by a client installed on the method.

Through the above steps S202 to S206, target point cloud data is obtained, and the target point cloud data is data obtained by scanning a target object around the vehicle through a laser wire harness; clustering the target point cloud data to obtain a plurality of first data sets, The feature points represented by the point cloud data included in each first data set are fitted to the same segmentation line segment, and the feature points are points on the target object; the plurality of first data are compared according to the distance between the plurality of segmentation line segments. The sets are merged to obtain a second data set, and the second data set includes at least one first data set. In the segmentation process, "clustering the target point cloud data to obtain multiple first data sets" is equivalent to completing the point cloud data segmentation only after traversing all the point cloud data once, instead of using multiple Iterating over the point cloud data to complete the segmentation can solve the technical problem of low efficiency of point cloud segmentation in related technologies, and then achieve the technical effect of improving the segmentation efficiency.

With the mass production of low-beam lidar, accurate segmentation of low-beam laser point cloud data is the basic prerequisite for obstacle detection and tracking, and it is of great significance in driverless perception technology. The fast segmentation method based on low-beam laser point cloud data proposed by the present application (also applicable to multi-beam lidar), while reducing the computational complexity, is helpful to reduce the over-segmentation and under-segmentation during the segmentation process, and improve the automatic The perception of driving is robust. The steps shown in FIG. 2 are combined below to detail the technical solution of the present application.

Optionally, in the technical solution provided in step S202, the technical solution of the present application may be applied to the field of autonomous driving. In order to enable the server to obtain target point cloud data, a laser sensor (or laser radar) may be installed on the vehicle. In order to reduce the cost, the lidar can be a low-beam lidar. In this way, when acquiring the target point cloud data, the laser point sensors installed on the vehicle can be used to scan the target objects around the vehicle to obtain the target point cloud data. The vehicle may be a vehicle having an automatic driving system.

In the technical solution provided in step S204, the server clusters the target point cloud data to obtain multiple first data sets, and the feature points represented by the point cloud data included in each first data set are fitted to the same segmentation line segment. The feature points are points on the target object.

In the above embodiment, when the target point cloud data is clustered to obtain multiple first data sets, each first data set may be created in the following manner (including steps 1 to 2):

Step 1: Find a plurality of first point cloud data in the target point cloud data, and the feature points represented by the plurality of first point cloud data are adjacent.

Optionally, finding multiple first point cloud data in the target point cloud data includes: setting the distance between the feature points represented in the target point cloud data to be not greater than a first threshold, and the distance between the represented feature points. The point cloud data with the included angle not less than the second threshold is regarded as a plurality of first point cloud data. An optional implementation is as follows:

Step 11: Obtain the second point cloud data in the target point cloud data, and the second point cloud data is the point cloud data in the target point cloud data that is not clustered into any one of the first data sets.

For example, point cloud data is sequentially acquired according to the point cloud data acquisition time in the target point cloud data. For example, acquiring the point cloud data in the order of morning to night (or late to early) in the acquisition time, in general, can become The characteristic parts (such as edges, surfaces, edges and corners) of a target object (such as obstacles) are often adjacent in position, and the lidar scans the target object in turn according to the position, in other words, the acquisition time phase Adjacent point cloud data is used to represent adjacent feature points. It can be seen that each first data set generally retains multiple second point cloud data adjacent to each other at the acquisition time. In other words, the above solution is to use continuous point clouds. The data is divided into multiple segments, and the point cloud data of each segment is stored in a first data set.

Optionally, the above-mentioned target point cloud data can also be fitted to obtain lines (that is, segmented line segments), and it can be defined that the distance between the lines does not exceed a certain threshold. The distance exceeds the threshold, then the two points can be respectively the endpoints of the line, and then multiple lines can be determined, and the point cloud data corresponding to all points on each line is used as a first data set.

Step 12. Obtain point cloud data (collected as third point cloud data) whose acquisition time is later than the second point cloud data.

The above-mentioned second point cloud data is equivalent to the starting point cloud data in a first data set. After this, the end point cloud data of the first data set needs to be found. The third point cloud data is point cloud data in the target point cloud data that has not been clustered into any of the first data sets and is collected later than the second point cloud data.

Step S13: Obtain the distance between the third point cloud data and the fourth point cloud data (that is, the point cloud data collected later than the third point cloud data and the adjacent point cloud data), and the first point cloud data represented by the third point cloud data. An angle formed between a feature point A, a second feature point B represented by the fourth point cloud data, and a third feature point C represented by the fifth point cloud data (that is, the angle of ∠ABC).

Step 14. If the distance between the feature points represented by the third point cloud data and the feature points represented by the fourth point cloud data is greater than a first threshold (the first threshold is used to characterize adjacent features of the same feature of the obstacle) The maximum distance between points, such as the maximum distance between pixels at the edge of a building), and the feature points represented by the third point cloud data, the feature points represented by the fourth point cloud data, and the fifth point cloud data. The included angle formed between the represented feature points is smaller than the second threshold (the second threshold is used to characterize the maximum corner between adjacent feature points of the same feature of the obstacle). The cloud data and the point cloud data whose acquisition time is between the acquisition time of the second point cloud data and the acquisition time of the third point cloud data are regarded as the plurality of first point cloud data.

In other words, the third point cloud data is equivalent to the end point cloud data of the first data set, the third point cloud data and the fourth point cloud data are point cloud data adjacent to the acquisition time, and the fourth point cloud data is collected late. At the time of the third point cloud data collection, the fourth point cloud data is equivalent to the starting point cloud data of the next first data set, and the fourth point cloud data and the fifth point cloud data are adjacent point cloud data at the time of collection and The collection time of the fifth point cloud data is later than the collection time of the fourth point cloud data.

Step 15: If the distance between the feature point represented by the third point cloud data and the feature point represented by the fourth point cloud data is greater than the first threshold, and the feature point represented by the third point cloud data and the fourth point cloud The angle formed between the feature points represented by the data and the feature points represented by the fifth point cloud data is less than the second threshold value, and the fourth point cloud data is saved to a different one from the one used to save the third point cloud data. The first data set.

Step 2. Save multiple first point cloud data to the same first data set created.

Subsequent point cloud data may be processed according to the foregoing steps 1 to 2 until point cloud data does not exist in the target point cloud data.

Through the technical solution provided in step S204, a preliminary segmentation of the point cloud data can be achieved.

In the technical solution provided in step S206, the server merges a plurality of first data sets according to the distance between the plurality of divided line segments to obtain a second data set, and the second data set includes at least one first data set.

In the above embodiment, a plurality of first data sets are combined according to the distance between the plurality of segmented line segments, and when a second data set is obtained, the interval between the segmented line segments obtained by the plurality of first data sets may be combined. The first data set with a distance less than the third threshold is merged to obtain a second data set. An optional implementation may include the following steps 1 to 2:

Step 1. Create an event collection.

In the event set, multiple segment line events corresponding to the multiple first data sets are stored according to the collection time of the point cloud data in the multiple first data sets. The segment line event includes an insertion event corresponding to the starting feature point of the split line segment. The deletion event corresponding to the ending feature point of the segmentation line segment.

Step 2. Iterate through each event in the event collection.

In step 3, when the traversed current event is an insertion event, the first divided line segment corresponding to the current event in the plurality of divided line segments is saved to a line segment set.

Optionally, since the point cloud data on each segment line segment is point cloud data with continuous acquisition time, then the point cloud data equivalent to each segment line segment can actually correspond to a collection period, so in the event collection, Events for dividing line segments are stored according to the collection time of the dividing line segments. For different dividing line segments, the collecting time is placed first at the head of the team, followed by the team head, and so on.

In step 4, if the current event is a delete event and the second segment line segment does not exist in the line segment set, the first data set corresponding to the first segment line segment is used as a third data set.

The second segmented line segment is that the distance between the line segment set and the first segmented line segment is less than a third threshold (the third threshold may be a parameter used to determine whether the two segmented line segments can be merged, and may be an empirical value or an experimental value. According to the environment at that time).

Step 5. In the case where the current event is a delete event and a second segment line segment exists in the line segment set, the first data set corresponding to the first segment line segment is merged into the first data set corresponding to the second segment line segment to obtain A third data set.

In other words, there may be a case of over-segmentation in the foregoing step S204. Therefore, the two first data sets that are over-segmented may be merged, and the corresponding divided line segments may also be merged into one.

Step 6. Determine the second data set according to the obtained multiple third data sets.

Optionally, determining the second data set according to the obtained multiple third data sets includes: directly using the multiple third data sets as the multiple second data sets. Or, perform denoising processing on multiple third data sets to obtain a second data set. For example, perform denoising processing on each third data set separately. If point cloud data still exists in the third data set after denoising processing, As a second dataset.

In the above embodiment, denoising processing may be performed on each third data set as follows:

First, if the number of point cloud data in the set is less than a specified threshold, then the point cloud data included in the set is more likely to belong to the noise rather than the target object. Therefore, in a possible implementation manner, the number of point cloud data in the third data set can be obtained. If the number of point cloud data in the third data set is less than the fourth threshold, the third data set is deleted. The set of culling point cloud data is less than the minimum point threshold value N _min (that is, the fourth threshold value), that is, the third data set excluding the point cloud data that may belong to noise.

Secondly, in some cases, the third data set may be a set of ground points obtained by scanning the ground points instead of the target object, and the ground point set will cause trouble for the classification, recognition and tracking of subsequent obstacles (target objects). Therefore, in a possible implementation manner, the distance between the center of gravity of the feature point represented by the point cloud data in the third data set and the laser sensor, and the number of scanning lines of the point cloud data in the third data set can be obtained. The distance from the laser sensor is less than the fifth threshold and the number of scan lines is less than 2. The third data set is deleted, that is, if the distance between the center of gravity of a certain data set and the sensor origin is less than a given distance threshold D _min (that is, the fifth threshold), and If the number of scanning lines is less than 2 layers, the data set can be considered as a set of ground points and removed.

As an optional embodiment, the technical solution of the present application is described in detail below by taking the technical solution of the present application as an example of automatic driving.

Autonomous vehicles (ie, Autonomous Vehicles or Self-piloting Automobiles), also known as self-driving cars, computer-driven cars, or wheeled mobile robots, are smart cars that realize driverlessness through computer systems. Autonomous cars rely on artificial intelligence , Visual computing, radar, monitoring devices, and global positioning systems work together to allow computers to operate motor vehicles automatically and safely without any human intervention.

In recent years, with the rise of driverless technology, the multi-wire beam lidar shown in Figure 3 has developed vigorously (64 wire beam lidar and 32 wire beam lidar are typical representatives of them). A solid black line represents a laser harness. Compared with single-line beam lidar, multi-line beam lidar can scan multiple scan lines at one time, and can quickly obtain rich three-dimensional information of the surrounding environment. It is very suitable for three-dimensional environment perception of unmanned driving systems. As shown in Figure 4, each A circle represents the point cloud data obtained by scanning a laser beam.

For the sake of cost, convenience of application, etc., the current research and development of lidar are moving towards miniaturization and low-wire beam development. Especially in recent years when ADAS applications have been implemented, low-wire beam laser radar has played an increasingly important role. The Level 3 autopilot system on the vehicle uses a 4-wire beam lidar, as shown in Figure 5.

Because a single sensor has its own drawbacks, multi-sensor fusion is often required to achieve robust environmental perception. In unmanned vehicle sensing systems, vision, millimeter wave and lidar data are often fused, compared to independent systems. , Which can make better and safer decisions, bad weather conditions or insufficient lighting conditions are not conducive to the camera, but the camera can distinguish colors (can identify traffic lights and street signs and other information), and has a high resolution; Lidar can accurately measure the distance information of surrounding obstacles, and it is not affected by ambient light. It can work normally at night.

There are many schemes for single-beam laser point cloud segmentation, but there are relatively few schemes for low-beam laser point cloud segmentation. They can be implemented by the following two technical schemes provided by this application: the first is to use the single-beam laser radar segmentation algorithm. Segmentation method based on the sequential arrangement of points within a single laser scanning line; the other is a method of segmenting 3D point cloud data directly, ignoring the continuity of the points.

The above method mainly has the following disadvantages: the segmentation method based on a single scan line only considers the distance and direction changes between adjacent points, it is easy to cause over-segmentation when the target is occluded, and it is easy to cause when the distance between target objects is too close Under-segmentation; In addition, segmented merges between different scan lines require multiple layers of loops, which are less efficient; a segmentation method based directly on 3D point cloud data, considering only the distance between points as a similarity measure, also exists Easy to over-segmentation and under-segmentation.

In order to overcome the above problems, the present application provides a fast segmentation method of a low-beam laser point cloud. This method first obtains segmentation segments on each scan line based on the segmentation method of each scan line; and then uses a planar scan method to perform each scan Line segmentation segments are merged to obtain candidate target clustering sets. Finally, feature extraction is performed on the candidate target clustering sets to remove noise and ground sets to obtain the final segmentation result. The basic flow of the low-beam laser point cloud (that is, target point cloud data) fast segmentation method is shown in Figure 6 below. This method is performed on a vehicle terminal as an example:

In step S602, the vehicle-mounted terminal performs segmentation of the scanning line of the low-beam laser point cloud layer by layer.

Referring to FIG. 7, a scanning line segmentation scheme considering distance and angular continuity is implemented as follows:

1) Initialize the segment segment set M to be empty, initialize lamda and Amax, and add the first point P1 (equivalent to the second point cloud data) to the first segment S ₁ (equivalent to a first data set) ;

2) Starting from the second point P ₂ , traverse each P _i on the scan line (equivalent to the fourth point cloud data), if || P _i -P _i-1 || (that is, the third point cloud data The distance between P _i-1 and the fourth point cloud data P _i ) is greater than D _max (the first threshold), and P _i-1 , P _i and P _{i + 1} (equivalent to the fifth point cloud data) are 3 The angle formed by the points is less than 180-A _max (equivalent to the second threshold), then the current segment S _{k is} saved in M, and a new segment S _{k + 1 is} created (equivalent to another first data set) and inserted into the current point P _i S _{k + 1;} otherwise, the current point P _i of S _k inserted;

3) After traversing all the points, if the current segment S _{n is} not empty, insert the current segment S _n into the set M.

Interruption and merging of scan line segmentation: If the scan line segmentation shows a non-convex shape, it needs to be interrupted at the turning point; if a certain two segmentation segments are blocked by the foreground object and are divided into two parts, you can proceed merge.

In step S604, the in-vehicle terminal merges the divided segments based on the plane scanning method.

The plane scanning algorithm is the basic algorithm in computational geometry. It is used to calculate the intersection points of several line segments on the plane. Here, the plane scanning algorithm is improved to realize the merging of the divided segments of the scanning lines. The specific algorithm flow is as follows:

1) Define event as the start or end endpoint of the segment, where the start point corresponds to the insertion event of the current segment, and the end point corresponds to the delete event of the current segment; define the event queue Q (equivalent to the event set) as all events Ordered set; defines the current segment set S;

2) Initialize the event queue Q and the current segment set S (equivalent to the line segment set) to be empty, and insert the endpoints of all the scan line segment segments into the event queue Q and sort them from small to large;

3) Traverse each event in the event queue Q in turn. If the event is an insert event, insert its corresponding segment (the first segment line segment) into the current segment set S; if the event is a delete event, then First calculate the distance between its corresponding segment and other segments in the current segment set S. If the distance is less than a given threshold D _thred (equivalent to the third threshold), it means that there is a second segment line segment, and the current segment set is set Merge into the nearest segment set; otherwise, output the current segment set; and finally delete its corresponding segment set from the current segment set S.

The method for calculating the distance between the divided segments may be to calculate an average distance in an overlapping area between the two divided segments.

Step S606: The vehicle-mounted terminal removes noise and ground set.

1) Noise point cloud set culling: the number of point cloud data is less than the minimum point threshold _Nmin (equivalent to the fourth threshold), because if the number of point cloud data is less than the specified threshold, it means that the point cloud There is a high probability that the data dataset belongs to noise; in addition, the number of point cloud data is too small to calculate related features;

2) Ground point cloud set culling: If the distance between the center of gravity of the point cloud set and the origin of the sensor is less than a given distance threshold D _min (equivalent to the fifth threshold), and the number of scan lines is less than 2 layers, the point cloud set can be considered as ground Point collection and removal.

Refer to Figure 7 for the schematic diagram of adaptive distance threshold calculation:

among them,

Is the angle between the point cloud data P _n-1 and the x-axis,

Is the angle between the point cloud data P _n and the x-axis, is

Yes

versus

The difference between them, D _max represents the radius with P _n-1 as the center,

Represents the intersection between the line passing through the origin and P _n and the circle, r _n-1 represents the line segment from the origin to P _n-1 , and λ represents the line passing through the origin and P _n-1 and

The angle between the line and P _n-1 , σ _r is a preset parameter, which can be an empirical value or an experimental value.

Based on the above steps, rapid segmentation of the low-beam laser point cloud can be achieved, and the final segmentation result is obtained, as shown in FIG. 8 at 801. The point cloud data belonging to different target objects in the segmentation result can be identified by different colors or by using different shape identification frames. 801 The point cloud data belonging to the same target object (such as a railing) is identified by a rectangular frame.

Aiming at the following problems in the related technology: the segmentation method based on a single scan line only considers the distance and direction changes between adjacent points, and it is easy to cause over-segmentation when the target object is occluded, and when the distance between the target objects is too close It is easy to cause under-segmentation. In addition, the segmentation of segmented lines between different scanning lines requires multiple layers of loops, which is relatively inefficient. The segmentation method based on 3D point cloud data directly considers only the distance between points as a similarity measure. There is also the problem of easy over-segmentation and under-segmentation. In the solution of the present application, a fast segmentation method of a low-beam laser point cloud is provided. This method firstly obtains the segmentation segments on each scan line based on the segmentation method of each scan line; then uses the planar scanning method to combine the segmentation segments of each scan line to obtain the candidate target cluster set; finally, the candidate target cluster set Feature extraction is performed to remove the ground and noise sets to obtain the final segmentation result.

With the mass production of low-beam lidar, accurate segmentation of low-beam laser point cloud data is the basic prerequisite for obstacle detection and tracking, and it is of great significance in driverless perception technology. The fast segmentation method based on low-beam laser point cloud data proposed in the present application, while reducing the complexity of the algorithm, is conducive to reducing the over-segmentation and under-segmentation during the segmentation process, and improves the perceived robustness of autonomous driving.

It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all described as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action order. Because according to the present application, certain steps may be performed in another order or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for this application.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary universal hardware platform, and of course, also by hardware, but in many cases the former is Better implementation. Based on such an understanding, the technical solution of this application that is essentially or contributes to the existing technology can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, The optical disc) includes several instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in the embodiments of the present application.

According to another aspect of the embodiments of the present application, a point cloud data segmentation device for implementing the above method for segmenting point cloud data is also provided. FIG. 9 is a schematic diagram of an optional point cloud data segmentation device according to an embodiment of the present application. As shown in FIG. 9, the device may include:

The obtaining unit 901 is configured to obtain target point cloud data, where the target point cloud data is data obtained by scanning a target object around a vehicle through a laser beam.

In an implementation manner, the above-mentioned target point cloud data may be data obtained by scanning multiple laser beams of a lidar. The lidar may measure the size and shape of an object by using a scanning technique, and the lidar may adopt a stability And a precision rotating motor, when the laser beam hits the polygonal gauge driven by the motor to form a scanning beam, because the polygonal gauge is located on the front focal plane of the scanning lens, and the laser beam is evenly rotated to the mirror The incident angle changes relatively continuously, so the reflection angle also changes continuously. Through the function of the scanning lens, a parallel and continuous top-to-bottom scanning line is formed to form the scanning line data, that is, a single-beam laser scans once Sequence of formed point clouds.

The lidar of the present application may be a low-beam lidar or a multi-beam lidar. A low-beam lidar scan can generate fewer line-beam scan lines at a time. A low-beam lidar generally includes a 4-beam and an 8-beam, mainly 2.5D lidar. The vertical field of view generally does not exceed 10 °; multi-line beam lidar (or 3D lidar) can generate multiple scan lines in one scan. Multi-line beam lidar generally includes 16 line beams, 32 line beams, 64 line beams, etc. 3D lidar and 2.5 The biggest difference between D lidar lies in the range of vertical field of view of lidar. The range of vertical field of view of 3D laser mine can reach 30 ° or even more than 40 °.

A clustering unit 903 is configured to cluster the target point cloud data to obtain multiple first data sets, wherein the feature points represented by the point cloud data included in each first data set are fitted on the same segmentation line segment The feature points are points on the target object.

A merging unit 905 is configured to merge a plurality of first data sets according to a distance between a plurality of divided line segments to obtain a second data set, where the second data set includes at least one first data set.

It should be noted that the obtaining unit 901 in this embodiment may be used to execute step S202 in the embodiment of the present application, and the clustering unit 903 in this embodiment may be used to execute step S204 in the embodiment of the present application. The merging unit 905 in the example may be used to execute step S206 in the embodiment of the present application.

It should be noted here that the examples and application scenarios implemented by the foregoing modules and corresponding steps are the same, but are not limited to the content disclosed in the foregoing embodiments. It should be noted that, as part of the device, the above modules can be run in a hardware environment as shown in FIG. 1, and can be implemented by software or hardware.

The applicant analyzed the related technology and realized that most point cloud segmentation methods in the related technology deal with unordered and discrete point clouds. Among the point cloud segmentation methods, the cluster segmentation method has low complexity, It is easy to implement and can be used for segmentation of spatial point clouds. However, due to the existence of ground point clouds in large outdoor scenes, it is difficult to effectively segment ground and non-ground objects using clustering methods.

In an optional embodiment, in the non-ground point cloud clustering and segmentation part, a clustering segmentation method based on a fixed threshold radius may be adopted. The selection of the threshold value has a great influence on the segmentation result. If the threshold value is too large, the separation distance Closer small objects may not be separated (that is, under segmentation). If the threshold is selected too small, objects with a large distance (such as buildings) may be divided into multiple clusters (that is, over segmentation occurs).

In the embodiment of the present application, first, based on the segmentation method of each scan line, segment line segments (may be referred to as segment segments) on each scan line are obtained, and each segment line segment corresponds to a first data set, which can implement object The initial segmentation of the contours; then the plane scanning method is used to merge the segmented segments of each scan line to obtain the candidate target clustering set (that is, the second data set), which is equivalent to merging the contour lines that belong to an object object. The point cloud data needs to be traversed once, which can solve the problem of inefficiency and difficulty in effectively dividing the ground and non-ground objects when only using clustering and segmentation. Since the solution of this application only needs to determine the distance between segmented line segments, compared to the point cloud Distance, no accidental factor for judging the distance between point clouds (some point clouds overlap in a certain plane dimension due to perspective, but actually do not overlap in space), and can also avoid clustering based on a fixed threshold radius The method directly segmented the point cloud under segmentation or over segmentation. It can be seen that, while reducing the complexity of the algorithm, the technical solution of the present application is beneficial to reduce the over-segmentation and under-segmentation in the segmentation process, and improve the perceived robustness of autonomous driving.

Through the above modules, target point cloud data is obtained. The target point cloud data is data obtained by scanning target objects around the vehicle through a laser beam; the target point cloud data is clustered to obtain a plurality of first data sets, each of which The feature points represented by the point cloud data included in the data set are fitted to the same segmentation line segment, and the feature points are points on the target object; multiple first data sets are merged according to the distance between the plurality of segmentation line segments, A second data set is obtained, and the second data set includes at least one first data set. In the segmentation process, "clustering the target point cloud data to obtain multiple first data sets" is equivalent to completing the point cloud data segmentation only after traversing all the point cloud data once, instead of using multiple Iterating over the point cloud data to complete the segmentation can solve the technical problem of low efficiency of point cloud segmentation in related technologies, and then achieve the technical effect of improving the segmentation efficiency.

Optionally, the above-mentioned clustering unit may include: a finding module for finding a plurality of first point cloud data in the target point cloud data, wherein the feature points represented by the plurality of first point cloud data are adjacent; the first A save module, configured to save multiple first point cloud data to the same first data set created.

The above search module may also be used as point cloud data in which the distance between the feature points represented in the target point cloud data is not greater than the first threshold and the included angle formed between the represented feature points is not less than the second threshold Multiple first point cloud data.

Optionally, the search module may include: an acquisition submodule, configured to acquire second point cloud data in the target point cloud data, where the second point cloud data is not clustered to any one of the first point cloud data Point cloud data of the data set; a search sub-module is used if the distance between the feature points represented by the third point cloud data and the feature points represented by the fourth point cloud data is greater than the first threshold, and the third point cloud data The included angle between the represented feature point, the feature point represented by the fourth point cloud data, and the feature point represented by the fifth point cloud data is smaller than the second threshold, and the second point cloud data and the third point cloud are The data and the point cloud data whose acquisition time is between the time when the second point cloud data is collected and the time when the third point cloud data is collected are regarded as a plurality of first point cloud data, wherein the third point cloud data is the target point cloud data. The point cloud data that has not been clustered into any of the first data sets, and the collection time of the third point cloud data is later than the collection time of the second point cloud data. The third point cloud data and the fourth point cloud data are the collection time. Adjacent point cloud data The point cloud data collection time is later than the third point cloud data collection time, the fourth point cloud data and the fifth point cloud data are adjacent point cloud data at the collection time and the fifth point cloud data collection time is later than the fourth Collection time of point cloud data.

Optionally, the clustering unit may further include: a second saving module, configured to: if the distance between the feature points represented by the third point cloud data and the feature points represented by the fourth point cloud data is greater than the first threshold, and The angle formed between the feature points represented by the third point cloud data, the feature points represented by the fourth point cloud data, and the feature points represented by the fifth point cloud data is less than the second threshold, and the fourth point cloud data is Save to another first data set than the one used to save the third point cloud data.

The above-mentioned merging unit may also be used for merging the first data set whose distance between the segmented line segments fitted in the plurality of first data sets is less than the third threshold to obtain the second data set.

Optionally, the merging unit may include: a creating module for creating an event set, wherein the event set stores multiple segment line segments corresponding to the multiple first data sets according to the collection time of the point cloud data in the multiple first data sets. The events of the segmented line segment include the insertion event corresponding to the starting feature point of the segmented line segment and the delete event corresponding to the end feature point of the segmented line segment; the merge module is used to traverse each event in the event set. When the current event of the is an insertion event, the first division line segment corresponding to the current event among the plurality of division line segments is saved to the line segment set; the case where the current event is a delete event and the second segment line segment does not exist in the line segment set In the following, the first data set corresponding to the first division line segment is taken as a third data set; if the current event is a delete event and a second division line segment exists in the line segment set, the first data set corresponding to the first division line segment will be A data set is merged into the first data set corresponding to the second segmentation line segment to obtain a third data set. The second segmentation line segment is The set of segments divided in the distance between the first line segment is less than the third threshold segmentation; determining means for determining a second set of data according to the third plurality of sets of data obtained.

Optionally, the foregoing determining module may be further configured to: use the plurality of third data sets as the second data set; and perform denoising processing on the plurality of third data sets to obtain the second data set.

The above determination module may be further configured to: obtain the number of point cloud data in the third data set; if the number of point cloud data in the third data set is less than the fourth threshold, delete the third data set; obtain the point cloud data in the third data set The distance between the center of gravity of the represented feature point and the laser sensor, and the number of scan lines of the point cloud data in the third data set. If the distance is less than the fifth threshold and the number of scan lines is less than 2, the third data set is deleted.

Optionally, the obtaining unit may be further configured to scan the target object by using a laser sensor installed on a vehicle to obtain target point cloud data, and the vehicle has an automatic driving system.

It should be noted here that the examples and application scenarios implemented by the foregoing modules and corresponding steps are the same, but are not limited to the content disclosed in the foregoing embodiments. It should be noted that, as part of the device, the above modules can be run in a hardware environment as shown in FIG. 1, and can be implemented by software or hardware, wherein the hardware environment includes a network environment.

According to another aspect of the embodiments of the present application, a server or terminal for implementing the method for segmenting point cloud data is also provided.

FIG. 10 is a structural block diagram of a terminal according to an embodiment of the present application. As shown in FIG. 10, the terminal may include: one or more processors (only one is shown in FIG. 10) a processor 1001, a memory 1003, and a transmission device. 1005. As shown in FIG. 10, the terminal may further include an input / output device 1007.

The memory 1003 may be used to store software programs and modules, such as program instructions / modules corresponding to the point cloud data segmentation method and device in the embodiments of the present application. The processor 1001 runs the software programs and modules stored in the memory 1003. Thus, various functional applications and data processing are performed, that is, the above-mentioned segmentation method of point cloud data is implemented. The memory 1003 may include a high-speed random access memory, and may further include a non-volatile memory, such as one or more magnetic storage devices, a flash memory, or other non-volatile solid-state memory. In some examples, the memory 1003 may further include a memory remotely set with respect to the processor 1001, and these remote memories may be connected to the terminal through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The transmission device 1005 is used to receive or send data through a network, and may also be used for data transmission between a processor and a memory. Specific examples of the foregoing network may include a wired network and a wireless network. In one example, the transmission device 1005 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers through a network cable so as to communicate with the Internet or a local area network. In one example, the transmission device 1005 is a radio frequency (RF) module, which is used to communicate with the Internet in a wireless manner.

Specifically, the memory 1003 is used to store an application program.

The processor 1001 may call the application program stored in the memory 1003 through the transmission device 1005 to perform the following steps:

Obtaining target point cloud data, where the target point cloud data is data obtained by scanning a target object around a vehicle with a laser beam;

Clustering the target point cloud data to obtain multiple first data sets, wherein the feature points represented by the point cloud data included in each first data set are fitted on the same segmentation line segment, and the feature points are on the target object Point

A plurality of first data sets are combined according to the distance between the plurality of division line segments to obtain a second data set, where each second data set includes at least one first data set.

Using the embodiment of the present application, target point cloud data is obtained, and the target point cloud data is data obtained by scanning a target object around the vehicle through a laser beam; the target point cloud data is clustered to obtain multiple first data sets, each The feature points represented by the point cloud data included in the first data set are fitted to the same segmentation line segment, and the feature points are points on the target object; multiple first data sets are performed according to the distance between the plurality of segmentation line segments. Combine to obtain a second data set, and the second data set includes at least one first data set. In the segmentation process, "clustering the target point cloud data to obtain multiple first data sets" is equivalent to completing the point cloud data segmentation only after traversing all the point cloud data once, instead of using multiple Iterating over the point cloud data to complete the segmentation can solve the technical problem of low efficiency of point cloud segmentation in related technologies, and then achieve the technical effect of improving the segmentation efficiency.

Optionally, for a specific example in this embodiment, reference may be made to the example described in the foregoing embodiment, which is not repeatedly described in this embodiment.

Those of ordinary skill in the art can understand that the structure shown in FIG. 10 is only for illustration, and the terminal may be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, and a mobile Internet device (Mobile Internet Devices) PAD and other terminal equipment. FIG. 10 does not limit the structure of the electronic device. For example, the terminal may also include more or fewer components (such as a network interface, a display device, etc.) than those shown in FIG. 10, or have a different configuration from that shown in FIG.

A person of ordinary skill in the art may understand that all or part of the steps in the various methods of the foregoing embodiments may be completed by using a program to instruct terminal-related hardware. The program may be stored in a computer-readable storage medium, and the storage medium may be Including: flash disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

An embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the foregoing storage medium may be used for program code for performing a method of segmenting point cloud data.

Optionally, in this embodiment, the storage medium may be located on at least one network device among multiple network devices in the network shown in the foregoing embodiment.

Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps:

S12. Obtain target point cloud data, where the target point cloud data is data obtained by scanning a target object around the vehicle through a laser beam;

S14. Clustering the target point cloud data to obtain multiple first data sets, where the feature points represented by the point cloud data included in each first data set are fitted on the same segmentation line segment, and the feature points are the target Point on object

S16. Combine multiple first data sets according to the distance between the multiple segmentation line segments to obtain a second data set, where the second data set includes at least one first data set.

Optionally, in this embodiment, the foregoing storage medium may include, but is not limited to, a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, and a magnetic disk. Various media such as discs or optical discs that can store program codes.

The above-mentioned serial numbers of the embodiments of the present application are merely for description, and do not represent the superiority or inferiority of the embodiments.

When the integrated unit in the foregoing embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in the computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially a part that contributes to the existing technology or all or part of the technical solution may be embodied in the form of a software product, which is stored in a storage medium. Several instructions are included to enable one or more computer devices (which may be personal computers, servers, or network devices, etc.) to perform all or part of the steps of the method described in the embodiments of the present application.

In the above embodiments of the present application, the description of each embodiment has its own emphasis. For a part that is not described in detail in an embodiment, reference may be made to the related description of other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the unit is only a logical function division. In actual implementation, there may be another division manner. For example, multiple units or components may be combined or may be combined. Integration into another system, or some features can be ignored or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, units or modules, and may be electrical or other forms.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, may be located in one place, or may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objective of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each of the units may exist separately physically, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional unit.

The above is only the preferred implementation of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principles of the present application, several improvements and retouches can be made. These improvements and retouches also It should be regarded as the protection scope of this application.

Claims

A method for segmenting point cloud data, the method being applied to a network device, including:

Acquiring target point cloud data, where the target point cloud data is data obtained by scanning a target object around a vehicle through a laser beam;

Clustering the target point cloud data to obtain a plurality of first data sets, wherein the feature points represented by the point cloud data included in each of the first data sets are fitted on the same segmentation line segment, the Feature points are points on the target object; and

Merging the plurality of first data sets according to the distance between the plurality of division line segments to obtain a second data set, wherein the second data set includes at least one of the first data sets.
The method of claim 1, wherein the clustering the target point cloud data to obtain a plurality of first data sets includes creating each of the first data sets as follows:

Find a plurality of first point cloud data in the target point cloud data, wherein the feature points represented by the plurality of first point cloud data are adjacent; and

Save the plurality of first point cloud data to the same first data set that is created.
The method according to claim 2, wherein finding a plurality of first point cloud data in the target point cloud data comprises:

And using the point cloud data in which the distance between the feature points represented in the target point cloud data is not greater than a first threshold and the included angle formed between the represented feature points is not less than a second threshold The first point is cloud data.
The method according to claim 3, wherein a distance between the feature points represented in the target point cloud data is not greater than a first threshold, and an included angle formed between the represented feature points is not smaller than the first The two threshold point cloud data as the plurality of first point cloud data include:

Acquiring second point cloud data in the target point cloud data, where the second point cloud data is point cloud data in the target point cloud data that is not clustered to any one of the first data sets; as well as

If the distance between the feature point represented by the third point cloud data and the feature point represented by the fourth point cloud data is greater than the first threshold, and the feature point represented by the third point cloud data, the first The angle formed between the feature points represented by the four point cloud data and the feature points represented by the fifth point cloud data is smaller than the second threshold, and the second point cloud data and the third point cloud data are divided. And the point cloud data whose acquisition time is between the acquisition time of the second point cloud data and the acquisition time of the third point cloud data is taken as the plurality of first point cloud data; wherein the third point cloud data; The data is point cloud data in the target point cloud data that has not been clustered into any of the first data sets, and the acquisition time of the third point cloud data is later than the acquisition time of the second point cloud data The third point cloud data and the fourth point cloud data are point cloud data whose acquisition time is adjacent and the acquisition time of the fourth point cloud data is later than the acquisition time of the third point cloud data, so The fourth point cloud data and the fifth point cloud data are acquisition time The acquisition time of the adjacent point cloud data and the fifth point cloud data is later than the acquisition time of the fourth point cloud data.
The method according to claim 4, wherein the method further comprises:

If the distance between the feature point represented by the third point cloud data and the feature point represented by the fourth point cloud data is greater than the first threshold, and the feature point represented by the third point cloud data The angle formed between the feature points represented by the fourth point cloud data and the feature points represented by the fifth point cloud data is smaller than the second threshold, and the fourth point cloud data is saved to Different from another said first data set for storing said third point cloud data.
The method according to any one of claims 1 to 5, wherein merging the plurality of first data sets according to a distance between the plurality of division line segments to obtain a second data set includes:

Merging the first data set whose distance between the segmented line segments fitted in the plurality of first data sets is less than a third threshold to obtain the second data set.
The method according to claim 6, wherein the first data set whose distance between the segmented line segments fitted in the plurality of first data sets is less than a third threshold is combined to obtain the The second data set includes:

An event set is created, wherein events in the plurality of first data sets corresponding to the plurality of segmented line segments are stored in the event set according to a collection time of point cloud data in the plurality of first data sets, and the segmentation The event of the line segment includes an insertion event corresponding to the start feature point of the segmented line segment and a delete event corresponding to the end feature point of the segmented line segment;

Traverse each event in the event set, and if the traversed current event is an insertion event, save a first segment line segment corresponding to the current event in the plurality of segment line segments to a segment set; If the current event is a delete event and a second segment line segment does not exist in the line segment set, the first data set corresponding to the first segment line segment is used as a third data set; If the current event is a delete event and the second segmented line segment exists in the segment set, the first data set corresponding to the first segmented line segment is merged to the one corresponding to the second segmented line segment. One third data set is obtained in the first data set, and the second segment line segment is a segment line segment in which the distance between the first segment line segment and the first segment line segment is less than a third threshold; and

Determining the second data set according to the obtained multiple third data sets.
The method according to claim 7, wherein determining the second data set according to the obtained multiple third data sets comprises:

Using a plurality of the third data sets as the second data set; or,

Performing denoising processing on a plurality of the third data sets to obtain the second data set.
The method according to claim 8, wherein performing denoising processing on the plurality of third data sets comprises:

Acquiring the number of point cloud data in the third data set;

If the number of point cloud data in the third data set is less than a fourth threshold, deleting the third data set;

and / or,

Acquiring the distance between the center of gravity of the feature point represented by the point cloud data in the third data set and the laser sensor, and the number of scanning lines of the point cloud data in the third data set; and

If the distance is less than a fifth threshold and the number of scan lines is less than 2, delete the third data set.
The method according to any one of claims 1 to 5, wherein obtaining target point cloud data comprises:

The target point cloud data is obtained by scanning the target object with a laser sensor installed on the vehicle; the vehicle has an automatic driving system.
A point cloud data segmentation device applied to network equipment includes:

An obtaining unit, configured to obtain target point cloud data, where the target point cloud data is data obtained by scanning a target object around a vehicle through a laser beam;

A clustering unit, configured to cluster the target point cloud data to obtain a plurality of first data sets, wherein feature points represented by point cloud data included in each of the first data sets are fitted to a same On the segmented line segment, the feature points are points on the target object; and

A merging unit, configured to merge the plurality of first data sets according to the distance between the plurality of division line segments to obtain a second data set, wherein the second data set includes at least one of the first data set.
The apparatus according to claim 11, wherein the clustering unit comprises:

A search module for searching a plurality of first point cloud data in the target point cloud data, wherein the feature points represented by the plurality of first point cloud data are adjacent; and

A first saving module is configured to save the plurality of first point cloud data to the same first data set that is created.
The apparatus according to claim 12, wherein the search module comprises:

An acquisition submodule, configured to acquire second point cloud data in the target point cloud data, where the second point cloud data is the first point data in the target point cloud data that is not clustered Collected point cloud data; and

A search sub-module, configured to: if the distance between the feature point represented by the third point cloud data and the feature point represented by the fourth point cloud data is greater than the first threshold, and The angle formed between the feature point, the feature point represented by the fourth point cloud data, and the feature point represented by the fifth point cloud data is less than the second threshold, and the second point cloud data, the The third point cloud data and the point cloud data whose acquisition time is between the acquisition time of the second point cloud data and the acquisition time of the third point cloud data are taken as the plurality of first point cloud data; The third point cloud data is point cloud data in the target point cloud data that is not clustered to any one of the first data sets, and the collection time of the third point cloud data is later than the second point cloud data. Acquisition time of point cloud data, the third point cloud data and the fourth point cloud data are point cloud data adjacent to the acquisition time and the fourth point cloud data is collected later than the third point cloud Data collection time, the fourth point cloud data and the fifth point The cloud data is point cloud data whose acquisition time is adjacent and the acquisition time of the fifth point cloud data is later than the acquisition time of the fourth point cloud data.
A non-transitory computer-readable storage medium, the storage medium including a stored program, wherein when the program runs, the method described in any one of claims 1 to 10 is executed.
An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes any one of claims 1 to 10 through the computer program. The method described.